Artificial Intelligence and Drug Discovery
Artificial Intelligence (AI) has become an integral part of our daily lives, from self-driving cars to email proofreading. In medicine, AI models are even used to design new molecules for drugs. However, understanding how AI makes these decisions can be just as challenging as understanding a human mind.
The Role of Explainable AI
This is where Explainable AI (XAI) comes in. XAI is a technology that helps clarify how AI models make decisions. Researchers are increasingly using XAI to scrutinize predictive AI models and understand their workings better, especially in fields like chemistry.
The Black Box Challenge in AI
AI models often function as "black boxes," meaning their internal processes are not visible. This lack of transparency is concerning when AI is used for critical applications like drug discovery. For instance, if an AI model predicts a molecule as a potential drug, but we can't see its reasoning, scientists and the public might be skeptical.
Seeking Justification and Transparency
"As scientists, we want justification," says Rebecca Davis, a chemistry professor. "Models that provide insight into AI decisions can make scientists more comfortable." XAI offers a solution by revealing the decision-making process of AI models.
Enhancing Drug Discovery with XAI
Rebecca Davis and her team are applying XAI to AI models used for drug discovery, focusing on predicting new antibiotic candidates. With antibiotic resistance on the rise, accurate and efficient prediction models are vital.
Unveiling AI's Hidden Insights
The team uses databases of known drug molecules to train AI models that predict a compound's biological effects. A collaborator, Pascal Friederich, developed an XAI model that identifies specific molecular parts influencing AI predictions. This approach helps researchers understand AI's criteria for classifying compounds.
Surprising Discoveries in Chemistry
XAI revealed surprising insights into penicillin molecules. While chemists focus on the penicillin core, XAI found that attached structures are crucial for antibiotic activity. "This explains why some penicillin derivatives show poor activity," notes Davis.
Toward Improved AI Models
The researchers hope to use XAI to refine predictive AI models further. "XAI shows us important factors for antibiotic activity," says Sturm, a graduate student. This knowledge can train AI models better.
Future Steps and Collaboration
Next, the team plans to collaborate with a microbiology lab to test compounds their refined models predict as effective antibiotics. They aim to develop better or new antibiotic compounds to combat resistant pathogens.
Building Trust in AI
"AI often causes distrust," says Davis. "But if AI can explain its actions, it will likely be more accepted." Sturm believes AI's role in chemistry and drug discovery is the future, and his work lays the groundwork for this transformation.
